How can classical multidimensional scaling go wrong?

Authors: Rishi Sonthalia, Greg Van Buskirk, Benjamin Raichel, Anna Gilbert

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we do two things. First, we empirically verify all of the theoretical claims. Second, we show that on the downstream task of classification, if we use c MDS to embed the data, then as the embedding dimension increases, the classification accuracy gets worse. Thus, suggesting that the embedding quality of c MDs degrades. [...] For the classification tasks, we switch to more standard benchmark datatsets: MNIST, Fashion MNIST, and CIFAR10.
Researcher Affiliation Academia Rishi Sonthalia University of Michigan rsonthal@umich.edu Gregory Van Buskirk University of Texas Dallas greg.vanbuskirk@utdallas.edu Benjamin Raichel University of Texas Dallas Benjamin.Raichel@utdallas.edu Anna C. Gilbert Yale University anna.gilbert@yale.edu
Pseudocode Yes Algorithm 1 Classical Multidimensional Scaling. [...] Algorithm 2 Lower Bound Algorithm.
Open Source Code No The paper does not provide concrete access to source code for the methodology described, such as a specific repository link, an explicit code release statement, or code in supplementary materials.
Open Datasets Yes First, are metrics that come from graphs and for these we use Celegans Rossi and Ahmed [2015] and Portugal Rozemberczki et al. [2019] datasets. [...] For both of these metrics we use the heart dataset Detrano et al. [1989]. [...] For the classification tasks, we switch to more standard benchmark datatsets: MNIST, Fashion MNIST, and CIFAR10.
Dataset Splits No The paper specifies a train/test split ('first 1000 images are training points' and 'tested the network on the remaining 1000 points') but does not mention a separate validation set or other specific data partitioning details like exact percentages for a three-way split.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup No The paper states 'trained a feed-forward 3 layer neural network' but does not provide specific experimental setup details such as hyperparameter values (e.g., learning rate, batch size, number of epochs) or optimizer settings.